Download the PHP package kch/rabbitmq-bundle without Composer

RabbitMqBundle

About

The RabbitMqBundle incorporates messaging in your application via RabbitMQ using the php-amqplib library.

The bundle implements several messaging patterns as seen on the Thumper library. Therefore publishing messages to RabbitMQ from a Symfony controller is as easy as:

Later when you want to consume 50 messages out of the upload_pictures queue, you just run on the CLI:

All the examples expect a running RabbitMQ server.

This bundle was presented at Symfony Live Paris 2011 conference. See the slides here.

Installation

For Symfony Framework >= 2.3

Require the bundle and its dependencies with composer:

Register the bundle:

Enjoy !

For a console application that uses Symfony Console, Dependency Injection and Config components

If you have a console application used to run RabbitMQ consumers, you do not need Symfony HttpKernel and FrameworkBundle. From version 1.6, you can use the Dependency Injection component to load this bundle configuration and services, and then use the consumer command.

Require the bundle in your composer.json file:

Register the extension and the compiler pass:

Warning - BC Breaking Changes

• Since 2012-06-04 Some default options for exchanges declared in the "producers" config section have changed to match the defaults of exchanges declared in the "consumers" section. The affected settings are:

  • durable was changed from false to true,
  • auto_delete was changed from true to false.

  Your configuration must be updated if you were relying on the previous default values.

• Since 2012-04-24 The ConsumerInterface::execute method signature has changed
• Since 2012-01-03 the consumers execute method gets the whole AMQP message object and not just the body. See the CHANGELOG file for more details.

Usage

Add the old_sound_rabbit_mq section in your configuration file:

Here we configure the connection service and the message endpoints that our application will have. In this example your service container will contain the service old_sound_rabbit_mq.upload_picture_producer and old_sound_rabbit_mq.upload_picture_consumer. The later expects that there's a service called upload_picture_service.

If you don't specify a connection for the client, the client will look for a connection with the same alias. So for our upload_picture the service container will look for an upload_picture connection.

If you need to add optional queue arguments, then your queue options can be something like this:

another example with message TTL of 20 seconds:

The argument value must be a list of datatype and value. Valid datatypes are:

• S - String
• I - Integer
• D - Decimal
• T - Timestamps
• F - Table
• A - Array

Adapt the arguments according to your needs.

If you want to bind queue with specific routing keys you can declare it in producer or consumer config:

Important notice - Lazy Connections

In a Symfony environment all services are fully bootstrapped for each request, from version >= 2.3 you can declare a service as lazy (Lazy Services). This bundle still doesn't support new Lazy Services feature but you can set lazy: true in your connection configuration to avoid unnecessary connections to your message broker in every request. It's extremely recommended to use lazy connections because performance reasons, nevertheless lazy option is disabled by default to avoid possible breaks in applications already using this bundle.

Import notice - Heartbeats

It's a good idea to set the to 2x the heartbeat so your socket will be open. If you don't do this, or use a different multiplier, there's a risk the consumer socket will timeout.

Dynamic Connection Parameters

Sometimes your connection information may need to be dynamic. Dynamic connection parameters allow you to supply or override parameters programmatically through a service.

e.g. In a scenario when the vhost parameter of the connection depends on the current tenant of your white-labeled application and you do not want (or can't) change it's configuration every time.

Define a service under connection_parameters_provider that implements the ConnectionParametersProviderInterface, and add it to the appropriate connections configuration.

Example Implementation:

In this case, the vhost parameter will be overridden by the output of getVhost().

Producers, Consumers, What?

In a messaging application, the process sending messages to the broker is called producer while the process receiving those messages is called consumer. In your application you will have several of them that you can list under their respective entries in the configuration.

Producer

A producer will be used to send messages to the server. In the AMQP Model, messages are sent to an exchange, this means that in the configuration for a producer you will have to specify the connection options along with the exchange options, which usually will be the name of the exchange and the type of it.

Now let's say that you want to process picture uploads in the background. After you move the picture to its final location, you will publish a message to server with the following information:

As you can see, if in your configuration you have a producer called upload_picture, then in the service container you will have a service called old_sound_rabbit_mq.upload_picture_producer.

Besides the message itself, the OldSound\RabbitMqBundle\RabbitMq\Producer#publish() method also accepts an optional routing key parameter and an optional array of additional properties. The array of additional properties allows you to alter the properties with which an PhpAmqpLib\Message\AMQPMessage object gets constructed by default. This way, for example, you can change the application headers.

You can use setContentType and setDeliveryMode methods in order to set the message content type and the message delivery mode respectively. Default values are text/plain for content type and 2 for delivery mode.

If you need to use a custom class for a producer (which should inherit from OldSound\RabbitMqBundle\RabbitMq\Producer), you can use the class option:

The next piece of the puzzle is to have a consumer that will take the message out of the queue and process it accordingly.

Consumers

A consumer will connect to the server and start a loop waiting for incoming messages to process. Depending on the specified callback for such consumer will be the behavior it will have. Let's review the consumer configuration from above:

As we see there, the callback option has a reference to an upload_picture_service. When the consumer gets a message from the server it will execute such callback. If for testing or debugging purposes you need to specify a different callback, then you can change it there.

Apart from the callback we also specify the connection to use, the same way as we do with a producer. The remaining options are the exchange_options and the queue_options. The exchange_options should be the same ones as those used for the producer. In the queue_options we will provide a queue name. Why?

As we said, messages in AMQP are published to an exchange. This doesn't mean the message has reached a queue. For this to happen, first we need to create such queue and then bind it to the exchange. The cool thing about this is that you can bind several queues to one exchange, in that way one message can arrive to several destinations. The advantage of this approach is the decoupling from the producer and the consumer. The producer does not care about how many consumers will process his messages. All it needs is that his message arrives to the server. In this way we can expand the actions we perform every time a picture is uploaded without the need to change code in our controller.

Now, how to run a consumer? There's a command for it that can be executed like this:

What does this mean? We are executing the upload_picture consumer telling it to consume only 50 messages. Every time the consumer receives a message from the server, it will execute the configured callback passing the AMQP message as an instance of the PhpAmqpLib\Message\AMQPMessage class. The message body can be obtained by calling $msg->body. By default the consumer will process messages in an endless loop for some definition of endless.

If you want to be sure that consumer will finish executing instantly on Unix signal, you can run command with flag -w.

Then the consumer will finish executing instantly.

For using command with this flag you need to install PHP with PCNTL extension.

If you want to establish a consumer memory limit, you can do it by using flag -l. In the following example, this flag adds 256 MB memory limit. Consumer will be stopped five MB before reaching 256MB in order to avoid a PHP Allowed memory size error.

If you want to remove all the messages awaiting in a queue, you can execute this command to purge this queue:

For deleting the consumer's queue, use this command:

Consumer Events

This can be useful in many scenarios. There are 3 AMQPEvents:

ON CONSUME

Let`s say you need to sleep / stop consumer/s on a new application deploy. You can listen for OnConsumeEvent (\OldSound\RabbitMqBundle\Event\OnConsumeEvent) and check for new application deploy.

BEFORE PROCESSING MESSAGE

Event raised before processing a AMQPMessage.

AFTER PROCESSING MESSAGE

Event raised after processing a AMQPMessage. If the process message will throw an Exception the event will not raise.

IDLE MESSAGE

Event raised when wait method exit by timeout without receiving a message. In order to make use of this event a consumer idle_timeout has to be configured. By default process exit on idle timeout, you can prevent it by setting $event->setForceStop(false) in a listener.

Idle timeout

If you need to set a timeout when there are no messages from your queue during a period of time, you can set the idle_timeout in seconds. The idle_timeout_exit_code specifies what exit code should be returned by the consumer when the idle timeout occurs. Without specifying it, the consumer will throw an PhpAmqpLib\Exception\AMQPTimeoutException exception.

Graceful max execution timeout

If you'd like your consumer to be running up to certain time and then gracefully exit, then set the graceful_max_execution.timeout in seconds. "Gracefully exit" means, that the consumer will exit either after the currently running task or immediatelly, when waiting for new tasks. The graceful_max_execution.exit_code specifies what exit code should be returned by the consumer when the graceful max execution timeout occurs. Without specifying it, the consumer will exit with status 0.

This feature is great in conjuction with supervisord, which together can allow for periodical memory leaks cleanup, connection with database/rabbitmq renewal and more.

Fair dispatching

You might have noticed that the dispatching still doesn't work exactly as we want. For example in a situation with two workers, when all odd messages are heavy and even messages are light, one worker will be constantly busy and the other one will do hardly any work. Well, RabbitMQ doesn't know anything about that and will still dispatch messages evenly.

This happens because RabbitMQ just dispatches a message when the message enters the queue. It doesn't look at the number of unacknowledged messages for a consumer. It just blindly dispatches every n-th message to the n-th consumer.

In order to defeat that we can use the basic.qos method with the prefetch_count=1 setting. This tells RabbitMQ not to give more than one message to a worker at a time. Or, in other words, don't dispatch a new message to a worker until it has processed and acknowledged the previous one. Instead, it will dispatch it to the next worker that is not still busy.

From: http://www.rabbitmq.com/tutorials/tutorial-two-python.html

Be careful as implementing the fair dispatching introduce a latency that will hurt performance (see this blogpost). But implemeting it allow you to scale horizontally dynamically as the queue is increasing. You should evaluate, as the blogpost recommends, the right value of prefetch_size accordingly with the time taken to process each message and your network performance.

With RabbitMqBundle, you can configure that qos_options per consumer like that:

Callbacks

Here's an example callback:

As you can see, this is as simple as implementing one method: ConsumerInterface::execute.

Keep in mind that your callbacks need to be registered as normal Symfony services. There you can inject the service container, the database service, the Symfony logger, and so on.

See https://github.com/php-amqplib/php-amqplib/blob/master/doc/AMQPMessage.md for more details of what's part of a message instance.

Recap

This seems to be quite a lot of work for just sending messages, let's recap to have a better overview. This is what we need to produce/consume messages:

• Add an entry for the consumer/producer in the configuration.
• Implement your callback.
• Start the consumer from the CLI.
• Add the code to publish messages inside the controller.

And that's it!

Audit / Logging

This was a requirement to have a traceability of messages received/published. In order to enable this you'll need to add "enable_logger" config to consumers or publishers.

If you would like you can also treat logging from queues with different handlers in monolog, by referencing channel "phpamqplib"

RPC or Reply/Response

So far we just have sent messages to consumers, but what if we want to get a reply from them? To achieve this we have to implement RPC calls into our application. This bundle makes it pretty easy to achieve such things with Symfony.

Let's add a RPC client and server into the configuration:

For a full configuration reference please use the php app/console config:dump-reference old_sound_rabbit_mq command.

Here we have a very useful server: it returns random integers to its clients. The callback used to process the request will be the random_int_server service. Now let's see how to invoke it from our controllers.

First we have to start the server from the command line:

And then add the following code to our controller:

As you can see there, if our client id is integer_store, then the service name will be __old_sound_rabbit_mq.integer_store_rpc__. Once we get that object we place a request on the server by calling addRequest that expects three parameters:

• The arguments to be sent to the remote procedure call.
• The name of the RPC server, in our case random_int.
• A request identifier for our call, in this case request_id.

The arguments we are sending are the min and max values for the rand() function. We send them by serializing an array. If our server expects JSON information, or XML, we will send such data here.

The final piece is to get the reply. Our PHP script will block till the server returns a value. The $replies variable will be an associative array where each reply from the server will contained in the respective request_id key.

By default the RPC Client expects the response to be serialized. If the server you are working with returns a non-serialized result then set the RPC client expect_serialized_response option to false. For example, if the integer_store server didn't serialize the result the client would be set as below:

You can also set a expiration for request in seconds, after which message will no longer be handled by server and client request will simply time out. Setting expiration for messages works only for RabbitMQ 3.x and above. Visit http://www.rabbitmq.com/ttl.html#per-message-ttl for more information.

As you can guess, we can also make parallel RPC calls.

Parallel RPC

Let's say that for rendering some webpage, you need to perform two database queries, one taking 5 seconds to complete and the other one taking 2 seconds –very expensive queries–. If you execute them sequentially, then your page will be ready to deliver in about 7 seconds. If you run them in parallel then you will have your page served in about 5 seconds. With RabbitMqBundle we can do such parallel calls with ease. Let's define a parallel client in the config and another RPC server:

Then this code should go in our controller:

Is very similar to the previous example, we just have an extra addRequest call. Also we provide meaningful request identifiers so later will be easier for us to find the reply we want in the $replies array.

Direct Reply-To clients

To enable direct reply-to clients you just have to enable option direct_reply_to on the __rpc_clients__ configuration for the client.

This option will use pseudo-queue amq.rabbitmq.reply-to when doing RPC calls. On the RPC server there is no modification needed.

Multiple Consumers

It's a good practice to have a lot of queues for logic separation. With a simple consumer you will have to create one worker (consumer) per queue and it can be hard to manage when dealing with many evolutions (forget to add a line in your supervisord configuration?). This is also useful for small queues as you may not want to have as many workers as queues, and want to regroup some tasks together without losing flexibility and separation principle.

Multiple consumers allow you to handle this use case by listening to multiple queues on the same consumer.

Here is how you can set a consumer with multiple queues:

The callback is now specified under each queues and must implement the ConsumerInterface like a simple consumer. All the options of queues-options in the consumer are available for each queue.

Be aware that all queues are under the same exchange, it's up to you to set the correct routing for callbacks.

The queues_provider is a optional service that dynamically provides queues. It must implement QueuesProviderInterface.

Be aware that queues providers are responsible for the proper calls to setDequeuer and that callbacks are callables (not ConsumerInterface). In case service providing queues implements DequeuerAwareInterface, a call to setDequeuer is added to the definition of the service with a DequeuerInterface currently being a MultipleConsumer.

Arbitrary Bindings

You may find that your application has a complex workflow and you you need to have arbitrary binding. Arbitrary binding scenarios might include exchange to exchange bindings via destination_is_exchange property.

The rabbitmq:setup-fabric command will declare exchanges and queues as defined in your producer, consumer and multi consumer configurations before it creates your arbitrary bindings. However, the rabbitmq:setup-fabric will NOT declare addition queues and exchanges defined in the bindings. It's up to you to make sure exchanges/queues are declared.

Dynamic Consumers

Sometimes you have to change the consumer's configuration on the fly. Dynamic consumers allow you to define the consumers queue options programmatically, based on the context.

e.g. In a scenario when the defined consumer must be responsible for a dynamic number of topics and you do not want (or can't) change it's configuration every time.

Define a service queue_options_provider that implements the QueueOptionsProviderInterface, and add it to your dynamic_consumers configuration.

Example Usage:

In this case the proc_logs consumer runs for server1 and it can decide over the queue options it uses.

Anonymous Consumers

Now, why will we ever need anonymous consumers? This sounds like some internet threat or something… Keep reading.

In AMQP there's a type of exchange called topic where the messages are routed to queues based on –you guess– the topic of the message. We can send logs about our application to a RabbiMQ topic exchange using as topic the hostname where the log was created and the severity of such log. The message body will be the log content and our routing keys the will be like this:

• server1.error
• server2.info
• server1.warning
• ...

Since we don't want to be filling up queues with unlimited logs what we can do is that when we want to monitor the system, we can launch a consumer that creates a queue and attaches to the logs exchange based on some topic, for example, we would like to see all the errors reported by our servers. The routing key will be something like: #.error. In such case we have to come up with a queue name, bind it to the exchange, get the logs, unbind it and delete the queue. Happily AMPQ provides a way to do this automatically if you provide the right options when you declare and bind the queue. The problem is that you don't want to remember all those options. For such reason we implemented the Anonymous Consumer pattern.

When we start an Anonymous Consumer, it will take care of such details and we just have to think about implementing the callback for when the messages arrive. Is it called Anonymous because it won't specify a queue name, but it will wait for RabbitMQ to assign a random one to it.

Now, how to configure and run such consumer?

There we specify the exchange name and it's type along with the callback that should be executed when a message arrives.

This Anonymous Consumer is now able to listen to Producers, which are linked to the same exchange and of type topic:

To start an Anonymous Consumer we use the following command:

The only new option compared to the commands that we have seen before is the one that specifies the routing key: -r '#.error'.

Batch Consumers

In some cases you will want to get a batch of messages and then do some processing on all of them. Batch consumers will allow you to define logic for this type of processing.

e.g: Imagine that you have a queue where you receive a message for inserting some information in the database, and you realize that if you do a batch insert is much better then by inserting one by one.

Define a callback service that implements BatchConsumerInterface and add the definition of the consumer to your configuration.

Note: If the keep_alive option is set to true, idle_timeout_exit_code will be ignored and the consumer process continues.

You can implement a batch consumer that will acknowledge all messages in one return or you can have control on what message to acknoledge.

How to run the following batch consumer:

Important: BatchConsumers will not have the -m|messages option available

STDIN Producer

There's a Command that reads data from STDIN and publishes it to a RabbitMQ queue. To use it first you have to configure a producer service in your configuration file like this:

That producer will publish messages to the words direct exchange. Of course you can adapt the configuration to whatever you like.

Then let's say you want to publish the contents of some XML files so they are processed by a farm of consumers. You could publish them by just using a command like this:

This means you can compose producers with plain Unix commands.

Let's decompose that one liner:

That command will find all the .xml files inside the symfony folder and will print the file name. Each of those file names is then piped to cat via xargs:

And finally the output of cat goes directly to our producer that is invoked like this:

It takes only one argument which is the name of the producer as you configured it in your config.yml file.

Other Commands

Setting up the RabbitMQ fabric

The purpose of this bundle is to let your application produce messages and publish them to some exchanges you configured.

In some cases and even if your configuration is right, the messages you are producing will not be routed to any queue because none exist. The consumer responsible for the queue consumption has to be run for the queue to be created.

Launching a command for each consumer can be a nightmare when the number of consumers is high.

In order to create exchanges, queues and bindings at once and be sure you will not lose any message, you can run the following command:

When desired, you can configure your consumers and producers to assume the RabbitMQ fabric is already defined. To do this, add the following to your configuration:

By default a consumer or producer will declare everything it needs with RabbitMQ when it starts. Be careful using this, when exchanges or queues are not defined, there will be errors. When you've changed any configuration you need to run the above setup-fabric command to declare your configuration.

How To Contribute

To contribute just open a Pull Request with your new code taking into account that if you add new features or modify existing ones you have to document in this README what they do. If you break BC then you have to document it as well. Also you have to update the CHANGELOG. So:

• Document New Features.
• Update CHANGELOG.
• Document BC breaking changes.

License

See: resources/meta/LICENSE.md

Credits

The bundle structure and the documentation is partially based on the RedisBundle